GAIA FLOW — Persistent project memory for AI coding agents

The problem

Context death is the tax on every AI-assisted project

The model is brilliant and amnesiac. The cost isn't the tokens — it's you, re-deriving the same decisions week after week.

Without project memory

Every session starts from zero — you paste context by hand
Decisions live in scrollback, then vanish
The same bug class returns because nobody wrote down the lesson
Two agents on one repo hold two different mental models
Come back after three weeks and you're a stranger in your own codebase

With GAIA FLOW

The agent reads the brain and is current in seconds
Every decision is recorded with its why, not just its what
Recurring errors become permanent lessons
One shared source of truth every agent resumes from
One click and you're back exactly where you stopped

How memory is organised

One folder. Structured so an agent can actually use it.

GAIA doesn't dump notes in a file. It builds a Project Brain — a layered memory where each document has one job, and reads in a defined order at the start of every session.

your-project/ ├── Project_Brain/ │ ├── 00_README.md what this is │ ├── 01_RESUME.md how to come back │ ├── 02_CURRENT_STATE.md live snapshot │ ├── 03_PHASE_TRACKER.md the roadmap │ ├── 04_QUICK_REFERENCE.md commands, ports │ ├── 05_ARCHITECTURE.md how it's built │ ├── 06_CONVENTIONS.md locked rules │ ├── PROJECT_MEMORY.md append-only log │ ├── DECISIONS/ why, not what │ └── INTELLIGENCE/ project DNA ├── RESUME.command one-click resume └── CLAUDE.md agent entry point

Evidence → Reasoning → Cognition

Deterministic scanners collect facts. The model reasons over those facts. The conclusion is persisted. Scripts never guess, and the model never asserts anything it can't cite.

Situational, not templated

A multi-site ops repo needs a site tracker. A mobile app needs a build plan. GAIA generates the documents the situation calls for — and keeps them current as the project moves.

The update protocol is locked

Finish a task, ship it to the brain. The snapshot and the changelog are never allowed to drift from what's actually true in the repo.

Mission control

Everything about the project, at a glance

A live terminal cockpit beside your agent: health, roadmap, risk, what the AI is doing right now, and what shipped last. Calm by design — it never steals focus and never slows the model down.

gaia cockpit — live

╭──────────────────────────────────────────────────────────────────────────────╮
│ 🧠 GAIA FLOW │ Acme Platform                                             │
│ phase P2.4   health ● clean   git main   mode ARCHITECTING            │
├──────────────────────────────────────────────────────────────────────────────┤
│ PROJECT HEALTH                                                               │
│   ✦ Production SaaS · Next.js + Postgres · 41k LOC · 12 services           │
│   next Split the billing service before the Q3 traffic ramp                │
│ Arch     ▓▓▓▓▓▓▓▓▓▓▓▓▓░░░  82%                                          │
│ Scale    ▓▓▓▓▓▓▓▓▓▓░░░░░░  64%                                          │
│ Risk     ▓▓▓▓▓░░░░░░░░░░░  31%                                          │
│ Maturity production                                                        │
│   trajectory scaling Moving from single-region to multi-tenant isolation    │
│   ◆ insight ● No rate limit on /api/export — three incidents cite it      │
│   ● stable · recovery ready                                               │
├──────────────────────────────────────────────────────────────────────────────┤
│ ✦ Refactoring the billing service… (4m 12s)                                │
│   ▰▰▰▰▰▰▰▰▰▰▰▱▱▱▱ 72%                                                  │
│   ■ Extract invoice generation into its own module                        │
│   □ Wire the new module into the payment orchestrator                     │
│   ✓ Write characterization tests for current behaviour                    │
│   … +4 completed                                                          │
├──────────────────────────────────────────────────────────────────────────────┤
│ ◆ AI ACTIVITY                                                               │
│   14:22:07 ❯ edit services/billing/invoice.ts                              │
│   14:22:31 ❯ bash pnpm test billing  ✓ 34 passed                        │
│   14:23:02 ❯ edit services/payments/orchestrator.ts ×3                    │
├──────────────────────────────────────────────────────────────────────────────┤
│ RECENT SHIPS                                                                 │
│   ✦ 2026-09-04 — Multi-tenant row-level security across 9 tables           │
│   ✦ 2026-09-02 — Decision #0031: Postgres over DynamoDB for the ledger     │
╰──────────────────────────────────────────────────────────────────────────────╯
   r refresh · q quit · live · GAIA FLOW · BRAHMA TECH

◐

Operational state

GAIA infers what you're actually doing — architecting, coding, debugging, deploying — and adapts what it surfaces to match.

▰

Live task intelligence

The current operation, real elapsed time and honest progress. The percentage is computed from real task state — never a fake timer.

◆

Proactive, but quiet

GAIA speaks only when it has something specific and evidenced to say. Silence is the default. Deep-work mode silences it entirely.

Using GAIA

Three commands is the whole workflow

Point GAIA at a project and it sets itself up — brain, launcher, cockpit, agent entry point. From then on, two commands carry every session.

STEP 01

Set the project up

Detects your stack, asks only what it can't observe, then scaffolds the brain and a one-click launcher for your OS.

Integrate GAIA FLOW

STEP 02

Start every session

The agent reads the brain in the right order and confirms it's in sync — no pasting, no re-explaining.

/gaia-resume

STEP 03

Ship the work into memory

Appends the changelog and refreshes the snapshot, so the next session opens already current.

/gaia-ship

Watch GAIA set up a project in 60 seconds

From empty folder to a project with a working memory.

▶

Demo video coming shortly

She gets smarter

GAIA learns how great engineering actually happens

GAIA studies real workflows — how problems get decomposed, how decisions get made, how agents plan, delegate and correct themselves. Those patterns compound into intelligence every project benefits from.

▸

Learns your patterns

How you structure work, verify it and ship it — so GAIA's suggestions read like your own senior engineer, not a generic checklist.

◈

Studies agentic work

How AI agents form plans, choose tools and recover from mistakes — captured as structured workflow traces.

✦

Compounds over time

Every project makes the engine sharper. What GAIA learns comes back as better protocols, better defaults, better advice.

Learning is opt-in and secret-scrubbed. Your source code is never uploaded.

About GAIA

Built for developers who live in the terminal

GAIA FLOW came out of a simple frustration: the AI was extraordinary at writing code and hopeless at remembering why any of it existed. Every session began with the same ritual of re-explanation.

So the fix wasn't a better prompt. It was memory that lives with the project — versioned in the repo, readable by any agent, and structured well enough to be trusted as the source of truth.

GAIA is deliberately terminal-first. No web app to alt-tab to, no dashboard to keep in sync. It runs where you already work, stays quiet while you're in flow, and speaks only when it has something specific to say.

Her purpose is to help developers build memory — so a project outlives the session that made it.

Zero hard dependencies

Pure shell at the floor. No Node, no Python required, no daemon in the background. It installs anywhere a terminal runs.

Never slows your agent

The highest law of the codebase. Expensive work is gated behind cheap change signals, so the model is never made to wait on GAIA.

Your code stays yours

The brain is plain markdown in your repo. No lock-in — delete the plugin and every note you have written is still there.

Your AI forgets everything.
GAIA remembers.